JP6411769B2 - Condition monitoring device - Google Patents

Description

  The present invention relates to a state monitoring device for monitoring the state of a machine.

  Work machines such as excavators and dump trucks used in mines and the like, and machines for social infrastructure such as gas turbines for power generation are required to operate 24 hours a day. In order to maintain high machine availability, unplanned machine stoppages must be prevented. For this purpose, it is necessary to shift from the regular maintenance based on the operation time of the conventional machine to the state monitoring maintenance that appropriately performs preventive maintenance based on the state of the machine. In order to realize state monitoring and maintenance, it is important to diagnose signs of machine abnormalities and failures by collecting and analyzing sensor data from various sensors provided in the machine by the state monitoring device. In order to improve the accuracy of diagnosis of this state monitoring device, it is important that the engineer regularly updates the diagnosis model so as to suppress false and misreported diagnosis. A false alarm is a case where the normal state of the machine is diagnosed as abnormal. Misreporting is a case where an abnormal state of the machine is diagnosed as normal. The diagnosis model is sensor data, a diagnosis method, and determination conditions used for diagnosis. For example, the determination condition includes a threshold value of target sensor data.

  As a technique for updating a diagnosis model in order to improve diagnosis accuracy, for example, there is (Patent Document 1). (Patent Document 1) discloses a method for creating a unified quality evaluation for a turbine mechanical system and the like and providing an automatic failure diagnosis tool. This content is: “Computer-implemented processes create and track machine unit signatures, machine site signatures, and machine fleet signatures to assess various operating events and provide fault detection. Sensor data collected from the system is transformed to correct or reduce data variability caused by ambient conditions and fuel quality, and the transformed data is analyzed using statistical methods to predict normal operating events. This information is used to create a single overall quality assessment of the event, by saving, tracking and further updating the operational event assessment over time, The technical content is that “deterioration of components is recognized at an arbitrary early stage”.

JP 2005-339558 A

  Consider a model that detects machine anomalies. (Patent Document 1) describes a method of setting a model parameter and a diagnosis threshold value for a machine alone to correct data variations caused by ambient conditions and fuel quality. That is, the update target item is fixed to the parameter and the threshold value. However, the method for improving the accuracy of diagnosis is not limited to setting parameters and thresholds. For example, when a model for detecting a new failure is added, a sensor to be used must be selected from a group of sensors of the machine. Furthermore, when applying an existing model to other machines, it is necessary to set the items separately for items that use the existing model in common and items that are individually set to eliminate individual differences between the machines. As described above, the items to be updated in the model change depending on the reason for adding and updating the model. When the model update target item is fixed as in (Patent Document 1), the model update work is aimed at improving the diagnostic accuracy of a model specialized for a specific machine, and the application is limited. Will be. In addition, when the model is freely updated, the user himself / herself has to determine an item to be updated of the model, which takes time to update the model.

  The present invention has been made in order to solve the above-described problems, and its purpose is to monitor a state that can reduce the man-hours for creating a high-accuracy diagnostic model while supporting the addition and update of the user's diagnostic model. Is to provide a device.

In order to achieve the above object, the present invention diagnoses a machine abnormality based on sensor data obtained by a plurality of sensors and a diagnostic model that is a condition for judging the machine abnormality, and displays the state of the machine. a condition monitoring apparatus, add or a trigger input which is activated when updating a storage unit that stores the diagnostic model of multiple machines, diagnostic model updating input from the trigger input section of the diagnostic model And a selection result of the additional reason, the original diagnosis model, a search unit for searching for similar cases based on the newly added diagnosis model, a diagnosis model searched by the search unit , a change table and a candidate diagnosis model a temporary storage unit for storing a display unit for displaying the diagnostic model and the candidate diagnosis model that reflects the changing table of the temporary storage unit, while viewing the display unit, additional Other inputs the changing of the diagnostic model updated user, an input section for selecting a candidate diagnostic model of the temporary storage unit, and the evaluation unit for verifying the new diagnostic model the candidate diagnosis model is reflected, An update unit for writing a new diagnostic model of the temporary storage unit into the storage unit based on the result of the evaluation unit is provided.

Furthermore, in the state monitoring apparatus of the present invention, the diagnosis model includes header information including a diagnosis model ID, sensor information, algorithm information, and model information including pre-processing information and post-processing information applied before and after the algorithm. And verification result information representing a result of verification using the diagnostic model described in the model information .

Furthermore, in the state monitoring apparatus of the present invention, the priority is given to the change part of the diagnosis model from the similar diagnosis model stored in the temporary storage unit, and the change is reflected in the change table stored in the temporary storage unit A partial determination unit is provided .

Furthermore, in the state monitoring apparatus of the present invention, a candidate generation unit that creates a diagnostic model candidate by combining information of a change unit of a similar diagnostic model using information of a change table stored in the temporary storage unit It is characterized by providing .

  According to the present invention, the user can reduce the man-hours of work by adding or updating a diagnostic model while browsing a similar diagnostic model created in the past.

  Furthermore, by displaying the fixed part and the changed part separately in the diagnostic model item, the user can save the trouble of determining the update item of the diagnostic model.

3 is a schematic diagram illustrating a flow of information of the state monitoring apparatus 100. 1 is a block diagram of a state monitoring device 100. FIG. It is a figure showing an example of the data format of the operation data storage part. It is a figure showing an example of the processing flow of the diagnostic part. It is a figure showing the outline | summary of the abnormality determination of a machine. It is a figure showing an example of the display of the diagnostic result display part. It is a figure showing an example of the screen of the trigger input part. 4 is a diagram illustrating an example of a data format of a storage unit 160. FIG. It is a figure showing an example of the data format of the temporary memory part. It is a figure showing an example of the processing flow of the change part determination part. It is a figure showing an example of the process of the change part determination part. It is a figure showing an example of the processing flow of the candidate production | generation part 190. FIG. 6 is a diagram illustrating a display example of a display unit 210. FIG. 6 is a diagram illustrating a display example of a display unit 210. FIG.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a diagram illustrating a flow of information among the machine 1 to be monitored, the state monitoring apparatus 100, the administrator 2, and the worker 3.

As shown in FIG. 1, the state monitoring device 100 periodically aggregates sensor data of the machine 1 via a wired or wireless communication system (not shown). Various sensors (not shown) are mounted on the machine 1. The manager 2 uses the state monitoring device 100 to monitor the state of the machine remotely from the machine 1. Furthermore, the administrator 2, when the state monitor 100 diagnoses the abnormality of the machine, contact the operator 3 who is in the field of machine 1, an instruction to perform maintenance of the machine 1.

  The state monitoring apparatus 100 has a screen that displays information related to the diagnosis result of the machine 1, and the administrator 2 monitors the state of the machine 1 while viewing this screen. For example, when a phenomenon in which abnormality frequently occurs in the diagnosis result of a specific machine 1, the administrator 2 determines whether or not the diagnosis model is appropriately set by looking at the screen. When the administrator 2 determines that the diagnostic model is not properly set, the administrator 2 sets the diagnostic model of the corresponding machine 1 again via the screen of the state monitoring apparatus 100. Then, the state monitoring apparatus 100 updates the stored diagnostic model to a newly set diagnostic model.

Next, FIG. 2 illustrates a detailed configuration of the state monitoring device 100 used in the state monitoring system according to the embodiment of the present invention. As shown in FIG. 2, the state monitoring apparatus 100 mainly includes an operation data storage unit 110, a diagnosis unit 120, a diagnosis result display unit 130, a trigger input unit 140, a search unit 150, a storage unit 160, a temporary storage unit 170, and a change. A part determination unit 180, a candidate generation unit 190, an evaluation unit 200, a display unit 210, an input unit 220, an update unit 230, and a user interface 240 are included.

The diagnosis unit 120, the diagnosis result display unit 130, the trigger input unit 140, the search unit 150, the changed part determination unit 180, the candidate generation unit 190, the evaluation unit 200, the display unit 210, the input unit 220, and the update unit 230 are a state monitoring device. 110 includes software processing means executed by a microprocessor, a RAM, a ROM (not shown), etc.

Operation data storage section 110, serial憶部160, the temporary storage unit 170, a hard disk, a recording device such as a flash memory. The user interface 240 has a display function such as a liquid crystal display and an input function such as a mouse and a keyboard. The user communicates with the diagnosis result display unit 130, the trigger input unit 140, the display unit 210, the input unit 220, and the update unit 230 via the user interface 240.

The operation data storage unit 110 stores various sensor data collected via the machine 1 and a wired or wireless communication system (not shown). The sensor data is collected at a predetermined timing. The operation data storage unit 110 stores sensor data of a plurality of types of machines. The data format of the operation data storage unit 110 is shown in FIG. The data format includes a machine table that defines the type of machine shown in FIG. 3A and a sensor table that stores sensor data of each machine shown in FIG. The items in the machine table are composed of a category, model, ID, and sensor table ID. These items are set by the user. The category represents the type of the target machine. For example, there are mining machines, turbines, and medical machines . The model represents the type of machine. The ID represents a category and model-specific ID. A machine is uniquely identified by a combination of model and ID. In FIG. 3A, two mining machines of the same model of model: A-ID: 001 and model: A-ID: 002 are defined. The sensor table ID is an ID for identifying the sensor table. The sensor table is a table that stores sensor data and is defined for each machine. FIG. 3B shows the format of the sensor table “KA001” of category: mining machine-model: A-ID: 001. The sensor table includes items of time and various sensors. This sensor item differs depending on the category and model.

  The diagnosis unit 120 uses the sensor data stored in the operation data storage unit 110 and the diagnosis model stored in the storage unit 160 to determine machine abnormality and provides the determination result to the diagnosis result display unit 130. To do. The diagnosis unit 120 is activated at a predetermined timing.

FIG. 4 shows a processing flow of the diagnosis unit 120. Step 101s treatment sandwiched S102, S103, S104, S105 in S101e (hereinafter, referred to as S101s) and all of the categories are stored in the machine table of the operation data storage unit 110 - ending process model -ID Repeat until. In step S102, the category-model-ID diagnosis model selected in step S101s is acquired from the storage unit 160. S104 and S105 sandwiched between S103s and S103e are repeated until the processing of all the diagnostic models acquired in S102 is completed. S104 is a process of determining an abnormality of the machine using the diagnosis model selected in S103s and the operation data of the category-model-ID selected in S101s. S105 is a process for collecting the diagnosis results of S104. As an example of the process of determining the machine abnormality in S104, there is a “K-average method” of multivariate analysis using a plurality of sensor data. The K-average method is a data classification method that classifies multivariate data without teaching data. By using this method, each input data is regarded as a point in the multivariate space, and a cluster of data can be found based on the closeness of the Euclidean distance of each point. Cluster information is included in the diagnostic model.

  FIG. 5 shows an example of abnormality determination using the K-average method in one of the diagnosis models of category: mining machine, model: A, ID: 001. In FIG. 5 (a), using trend data (2013/7/1 00:00:00 to 2013/7/1 23:59) of load factor, temperature, and pressure, and cluster information described in the diagnostic model, K Diagnose abnormalities with the average method. Specifically, neighboring clusters are found in the order of time of load factor, temperature, and pressure, and the Euclidean distance is calculated. In the K-average method, the Euclidean distance is the degree of abnormality. The trend data of the degree of abnormality is shown in FIG. If the degree of abnormality exceeds the threshold set in the diagnostic model, it is determined that the abnormality is a machine. In S105, the time when the threshold is exceeded, the average value of the degree of abnormality, and the duration that exceeds the threshold are calculated.

  The diagnosis result display unit 130 displays the diagnosis result provided from the diagnosis unit 120 on the screen of the user interface 240.

  A display example is shown in FIG. FIG. 6A shows a list of machine states of the display shown in the user interface 240. Specifically, for each machine, items of category, model, ID, failure mode, time, degree of abnormality, and duration, a detail button 6a1, an update button 6a2, and an add button 6a3 are displayed. The category, model, and ID are the same as the items defined in the machine table of the operation data storage unit 110. The failure mode is the same as the item of the diagnostic model item in the storage unit 160. The time, the degree of abnormality, and the duration are items calculated in S105 of the diagnosis unit 120. When the detail button is pressed, as shown in FIG. 6B, the trend data of the degree of abnormality calculated by the diagnosis unit 120 and the trend data of the sensor during the diagnosed period are displayed. Further, the position of the time determined to be abnormal is highlighted and displayed with an arrow 6b1. This is a function for quickly searching a time determined to be abnormal from a large amount of trend data. FIG. 6B shows an example when a detailed button of category: mining machine, model: A, ID: 001, failure mode: component A failure, time: 2013/7/1 19:00 is pressed. The arrow 6b1 points to the first time 2013/7/1 19:00 when the abnormality is detected. Also, the abnormality degree threshold value 6b2 defined in the diagnosis model is also displayed in the abnormality degree trend graph. If the user is not satisfied with the abnormality determination result, the update button 6a2 is pressed, and the screen changes to a screen for updating the diagnostic model. When a new failure mode or a new machine is desired to be added, the add button 6a3 is pressed, and the screen changes to a screen for adding a diagnostic model based on the corresponding diagnostic model. This diagnostic model update screen will be described in the paragraph describing the trigger input unit 140.

  The trigger input unit 140 is activated when a diagnostic model is added or updated. There are two timings for applying the update trigger. One is when the update button 6a2 of the diagnosis result display unit 130 in FIG. 6 is pressed. This is intended to improve diagnostic accuracy by pressing when the administrator is not satisfied with the determination result and updating the diagnostic model. The other is when the add button 6a3 of the diagnosis result display unit 130 of FIG. 6 is pressed. This aims to add a new failure mode, a new machine.

FIG. 7 shows a screen of the trigger input unit 140. FIG. 7A shows a screen when the update button 6a2 in FIG. 6 is pressed. FIG. 7A also displays a window 7a1 that displays the original diagnostic model of the display shown by the user interface 240. When updating, the reason why the diagnostic model is updated is selected from the selection screen. FIG. 7A includes “many false alarms”, “many false alarms”, and “others”. The false alarm is a case where the machine is determined to be abnormal although it is in a normal state. The unreported case is a case where the machine is determined to be normal although the machine is in an abnormal state. Improving the accuracy of a diagnostic model means bringing the number of false alarms and misreports close to zero. For reasons other than “Many false alarms” and “Many false alarms”, select “Other”. FIG. 7B shows a screen when the add button 6a3 in FIG. 6 is pressed. Here, the reason for the addition is selected from the selection screen 7b2. In FIG. 7B, “I want to add a failure mode”, “I want to add a new ID diagnostic model with the same category and model”, “I want to add a diagnostic model with the same category, new model, and ID”, “New I want to add a diagnostic model of category, model, ID. “I want to add a failure mode” is a case where a diagnosis model corresponding to a new failure mode is added to machines of the same category, model, and ID. “I want to add a diagnostic model with a new ID with the same category and model” is a case where a diagnostic model with a different ID is added to machines of the same category and model. “I want to add a diagnostic model of the same category, new model, and ID” is a case where I want to add a diagnostic model of a machine of the same category, new model, and ID. “ I want to add a new category, model, ID diagnosis model” is a case where I want to add a new category, model, ID diagnosis model.

On the display screen shown by the user interface 240 in FIG. 7B, a box 7b3 for inputting information of a newly added diagnostic model and a window 7b1 for displaying the original diagnostic model are also displayed. The information of the diagnostic model to be newly added is information regarding the failure mode, category, model, and ID to be added. User inputs the information of the diagnostic model to be newly added to the box 7b 3. The list of diagnostic model updates and addition reasons in FIG. 7 is predefined by the user. The trigger input unit 140 transmits information on the selected reason, the original diagnostic model, and the newly added diagnostic model to the search unit 150.

  The search unit 150 searches the storage unit 160 for similar cases by using the diagnosis model update and addition reason selection results input from the trigger input unit 140, the original diagnosis model, and the information of the newly added diagnosis model. The search result is stored in the temporary storage unit 170. Searches for similar cases include the same category, the same category and model, the same failure mode, and the same reason for updating and adding a diagnostic model. In addition to the complete match search, a search method such as a forward match search or a backward match search may be used. By using the diagnostic model created in the past in this way, it is possible to reduce the man-hours for updating and adding the diagnostic model.

  The storage unit 160 stores a diagnostic model. The storage unit 160 provides the diagnosis unit 120 with a diagnosis model of the target machine. Also, the search unit 150 becomes a search target. A new diagnostic model is written from the updating unit 230.

  FIG. 8 shows the data format of the storage unit 160. The diagnostic model includes header information shown in FIG. 8A, model information shown in FIG. 8B, verification result information shown in FIG. 8C, and update information shown in FIG. 8D.

  The header information shown in FIG. 8A includes a diagnosis model ID, a category, a machine number, an ID, and a failure mode. The category, number, and ID are information stored in the operation data storage unit 110. The diagnosis model ID is an ID for identifying a diagnosis model. The failure mode is the name of the failure to be detected by the diagnostic model or information on the failure site. One diagnostic model corresponds to one failure mode.

  The model information shown in FIG. 8B is information regarding the contents of the diagnostic model. In FIG. 8B, model information of failure of category: mining machine, number machine: A, ID: 001, failure mode: component A is described. The model information includes sensor information, pre-processing information, algorithm information, and post-processing information. The sensor information defines the sensor name used for the diagnostic model. Here, the load factor, pressure, and temperature are described. There are two types of preprocessing information: sensor conversion processing and state separation conditions. Sensor conversion processing refers to processing sensor data before a diagnostic algorithm. For example, if the sensor data contains a lot of noise, a moving average is applied. FIG. 8B shows information on processing for applying a moving average to load factor data. The state separation condition is a condition that defines the state of the machine. The state of the machine is divided into a steady state where the machine is moving stably and a transient state before the steady state is reached. For example, the engine is in a transient state that does not operate stably because the engine is not warmed immediately after startup, but becomes steady after a certain period of time. When diagnosis is performed for all machine states, there is a possibility that misdiagnosis of diagnosis may increase in a transient state. For this reason, it is possible to improve accuracy by separating and diagnosing the steady state of the machine in advance using sensor data. This extraction of the steady state of the machine is called state separation. For state separation, it is necessary to determine a sensor and its conditions for state separation. A sensor for separating a state and its condition are defined as a state separation condition. This state separation condition is also a part of the diagnostic model.

For example, in order to extract the steady state of the engine, a state separation condition in which the temperature sensor of the engine oil is 60 degrees or more is employed. This time, state separation conditions are not described. The algorithm information defines the name of the algorithm for determining abnormality and its parameter information. In the example of the diagnosis unit 120, the algorithm name is “K-average method”, and the parameter information is cluster information. The cluster information is managed as a file, and is described as Datafile0 in FIG. 8B. In the case of abnormality determination using principal component analysis, the algorithm name is “principal component analysis”, and the parameter information is information regarding the principal component. Post-processing information is processing for determining an abnormality after applying a diagnostic algorithm. For example, in the example of the diagnosis unit 120, the Euclidean distance from the cluster is defined as the degree of abnormality, and the machine abnormality is determined by comparing with the threshold value. If the threshold is 3, “abnormality level 3 or higher” is described in the post-processing information. In the post-processing, a threshold value and a duration time may be combined. For example, when the condition of “continuation of threshold value 3 or higher for 1 minute” is set, if the abnormality degree 3 continues for 1 minute, the machine is determined to be abnormal.

The verification result information shown in FIG. 8C is information on the result of verification using the diagnostic model described in the model information. The verification result information includes a learning period, a diagnosis period, the number of false reports, and the number of missed reports. The learning period defines a learning period when the algorithm of the algorithm information is machine learning. FIG. 8C describes the learning period 2013/6/1 00:00:00 to 2013/6/1 23:59. The diagnosis period defines the diagnosis period when the number of false reports and the number of false reports are calculated. FIG. 8C describes the diagnostic period 2013/7/1 00:00 to 2013/7/1 23:59. At this time, there are 3 false alarms and 0 false alarms. With this verification result information, the performance of the diagnostic model can be confirmed.

  The update information shown in FIG. 8D is information on the update history of the diagnostic model. The update history information includes version information, previous diagnosis model ID information, and reason information. The version information represents the version of the diagnostic model. The rule for adding this version is defined by the user. The previous diagnosis model ID information is information of the original diagnosis model based on which the diagnosis model is created. This information includes a diagnosis model ID that is header information. By following the previous diagnosis model ID, it is possible to refer to the evolution process of the diagnosis model. This is used by the changed part determination unit 180.

  The temporary storage unit 170 stores the search result of the search unit 150, the diagnostic model fixing unit and the change unit change table of the changed part determination unit 180, the diagnostic model candidate of the candidate generation unit 190, and the evaluation result of the evaluation unit 200. FIG. 9 shows the data format at this time. The diagnostic model temporary format shown in FIG. 9A is obtained by adding a temporary ID to the data format of the diagnostic model shown in FIG. The diagnostic model temporary format is used for the search result of the search unit 150 and the diagnostic model candidate of the candidate generation unit 190. The temporary ID is an ID indicating which processing unit has input the temporary ID. When input from the search unit 150, the temporary ID is 150. When input from the candidate generation unit 190, the temporary ID is 190. The evaluation result of the evaluation unit 200 is stored in the verification result information shown in FIG. 8C stored in the temporary storage unit 170. The change table shown in FIG. 9B is a table that defines a fixed part and a change part of the diagnostic model. The change part of the diagnostic model is described as “1”, and the fixed part is described as “0”. This is changed according to the reason for adding or updating the diagnostic model input from the trigger input unit 140. The predetermined diagnosis model change table is as follows. When the reason is “many false alarms” or “many false alarms”, the model information sensor, pre-processing, algorithm, post-processing, and learning period are changed, and the others are fixed.

  FIG. 9B shows this example. When “I want to add a failure mode”, the failure mode, the entire model information, and the entire verification result information are updated. When "I want to add a new ID diagnosis model with the same category and model", the category and model become fixed parts, and the others become change parts. When “I want to add a diagnosis model of the same category, new model, ID”, the category is fixed, and the others are changed. In the case of “I want to add a new category, model, ID diagnosis model”, all items are changed. It is also possible to calculate the priority of the change unit by the change part determination unit 180 and reflect it in the change table. In this case, the numerical value of the change part becomes larger as the priority is higher. For example, when there are many false alarms and the changed part determining unit 180 determines that the effect of the post-processing is large, the numerical value of the change part of the post-processing is set larger than other items.

  The data format of the candidate diagnosis model shown in FIG. 9C includes information on the changing unit, detailed information on the changing unit, and evaluation result information. FIG. 9C shows a case where the post-processing is a changing unit. The information of the change part is described as “post-processing”. Furthermore, post-processing of the candidate diagnosis model is described in the detailed information of the changing unit. Here, it is information on the threshold of the degree of abnormality. Furthermore, in the evaluation result information, the evaluation result of the diagnostic model when the post-processing of detailed information is reflected is displayed. The evaluation results at this time are the same as the verification result information items in the diagnostic model data format of FIG. 8 such as the number of false reports and the number of false reports.

  The changed part determining unit 180 gives priority to the changing part of the diagnostic model from the similar diagnostic models stored in the temporary storage unit 170 and reflects the priority in the change table stored in the temporary storage unit 170. FIG. 10 shows a processing flow of the changed part determining unit 180. S202, S203, and S204 sandwiched between S201s and S201e are repeatedly processed for the similar diagnosis model stored in the temporary storage unit 170. In S202, using the history information of the diagnostic model selected in S201, the diagnostic models for the past N times are acquired from the similar diagnostic models stored in the temporary storage 170 via the search unit 150. N is defined in advance by the user. When N is 0, the diagnostic model stored in the temporary storage unit 170 is indicated. S204 sandwiched between S203s and S203e is repeatedly calculated by going back to the past N diagnostic models. In step S203s, a difference between the past N-th diagnosis model and the past N + 1-th diagnosis model is calculated and a change item is specified. Further, in S204, the improvement rate α (n) is calculated with reference to the verification result information of the past Nth diagnosis model and the past N + 1th diagnosis model. For example, the improvement rate α (n) is calculated as follows.

α (n) = (the sum of the number of misreports and the number of misreports of the verification result information of the (N + 1) th diagnosis model) / (the sum of the number of misreports and the number of misreports of the Nth diagnosis model of the verification result information)
According to this definition, the greater the improvement rate, the higher the effect of changing the diagnostic item. In S205, the similar diagnosis model calculated so far and the improvement rate α (n) are totaled, and the effects of the changing unit are totaled. Specifically, change items and improvement rates are handled as a set, and the sum of the improvement rates is calculated for each same change item. The sum of the improvement rates becomes the priority of the change table.

  FIG. 11 shows an example of processing of the changed part determination unit 180. Items changed from the (N + 1) -th diagnosis model to the N-th diagnosis model are parameters for post-processing. Specifically, the threshold value was changed from 10 to 20. With this change, the number of false alarms has been improved from 100 to 10. The improvement rate at this time is 10. As described above, change items and improvement rates are handled as a set, and the priority of the change table is obtained by aggregation.

The candidate generation unit 190 uses the information of the change table stored in the temporary storage unit 170 to combine the information of the change units of similar diagnosis models to create a diagnosis model candidate. FIG. 12 shows a processing flow of the candidate generation unit 190. In step S301, a change unit with a high priority is extracted from the change table in the temporary storage unit 170. Here, the change unit may be extracted using a predetermined priority threshold value, or may be extracted using a predetermined number of change units. S303, S304, and S305 sandwiched between S302s and S302e are repeated for the number of changed parts extracted in S301. In step S303, a plurality of candidates are created by combining items of similar diagnosis models stored in the temporary storage unit 170 for the changed portion. For example, when the changing unit is a post-processing abnormality degree threshold value, the post-processing abnormality degree threshold value of a similar diagnosis model is extracted and reflected in the post-processing of the diagnostic model to be updated. When there are a plurality of similar diagnosis models, a plurality of candidate diagnosis models can be created. The change part of the change table changes according to the reason for updating the diagnostic model and the priority with which it is effective. For all the combinations, the candidate diagnosis model is not created, but only the changing unit is replaced to create the candidate diagnosis model, so that the number of combinations can be reduced. Thereby, calculation time can be shortened, and the update of a diagnostic model and an additional man-hour can be reduced. In S304, the evaluation unit 200 evaluates the candidate diagnosis model created in S303. S304 deletes a candidate diagnostic model, when the item of the candidate diagnostic model created in S303 and the diagnostic model update and addition target machine do not match. For example, it is assumed that the changing unit is a sensor, and the sensor of the “atmospheric temperature” is included in the sensors of the candidate diagnosis model. However, the target machine does not have a sensor for measuring the atmospheric temperature. In this case, the candidate diagnosis model presenting the sensor of “atmospheric temperature” is deleted. In S305, information on the candidate diagnosis model and information on the evaluation result are written in the data format of the candidate diagnosis model shown in FIG.

The evaluation unit 200 evaluates the performance of the candidate diagnosis model created by the candidate generation unit 190. Specifically, the performance of the candidate diagnostic model is evaluated using the diagnostic model update and the learning period and diagnostic period of the diagnostic model to be added. In the case of updating a diagnostic model, the number of false or missed reports is evaluated using the learning period and the diagnostic period of the verification result information of the data format of the diagnostic model to be updated. This process is the same as the diagnosis unit 120. In the case of adding a diagnostic model, since there is no diagnostic model as a base, the user inputs and sets a learning period and a diagnostic period.

  The display unit 210 displays the diagnostic model reflecting the change table in the temporary storage unit 170 and the candidate diagnostic model. FIG. 13 shows a display example of the display unit 210. FIG. 13A is a display example that reflects the information of the change table for the diagnostic model to be updated. The fixed part of the change table is dyed black and cannot be input by the input unit 220. In FIG. 13B, the changing units of the change table assign priorities (1 to 5) according to the priorities of the changing units. In FIG. 13B, the priority item is ranked 13b1 according to the priority of the change table. The emphasis display in FIG. 13 is not limited to one, which can be displayed by blinking, displaying larger than the surrounding characters, in addition to the method of coloring and emphasizing and the method of underlining. Absent.

  FIG. 14A shows a screen on which the user inputs a new diagnostic model via the input unit 220. FIG. 14B shows candidate diagnosis models in the temporary storage unit 170. At this time, candidate diagnosis models are displayed in order of effectiveness. This order is the order in which the verification results of the candidate diagnosis models in the temporary storage unit 170 are excellent. When a candidate diagnostic model is selected by the input unit 220, the candidate diagnostic model is reflected in the updated diagnostic model. The evaluation button 14a1 in FIG. 14A is a button for verifying a new diagnostic model input by the user. When this button is pressed, a new diagnostic model is verified using the evaluation unit 200 and is reflected in the verification result information in FIG. The update button 14a2 in FIG. 14A is a button for writing a new diagnostic model in the storage unit 160 when the user is satisfied with the performance of the new diagnostic model. This writing is performed via the update unit 230. FIG. 14 (a) highlights and displays the difference of the original diagnostic model. For example, FIG. 14A shows an example in which post-processing information is changed. At this time, the post-processing information is underlined in order to highlight and display that the post-processing information has been changed. FIG. 14C displays the similar diagnosis model stored in the temporary storage unit 170. The user can add or update the diagnostic model while referring to the similar diagnostic model, candidate diagnostic model, and diagnostic model change unit.

The update unit 230 writes a new diagnostic model in the storage unit 160 when the update button on the display unit 210 is pressed. At this time, the history information version of the data format of the diagnostic model is added. Specifically, if there is an original diagnostic model, a number is added to the version of the diagnostic model. The previous diagnosis model ID is also added. Diagnostic model update reason entered in the trigger input unit 140 further also stored. In new cases where there is no original diagnostic model, a new version is added. Further, the previous diagnosis model ID is not added. Diagnostic model update reason entered in the trigger input unit 140 further also stored.

  ADVANTAGE OF THE INVENTION According to this invention, referring to the diagnostic model produced in the past, a user's diagnostic model update and addition work can be supported, and it can implement | achieve reduction of an operation man-hour.

DESCRIPTION OF SYMBOLS 1 Machine 2 Manager 3 Worker 100 State monitoring apparatus 110 Operation data storage part 120 Diagnosis part 130 Diagnosis result display part 140 Trigger input part 150 Search part 160 Storage part 170 Temporary storage part 180 Change part determination part 190 Candidate generation part 200 Evaluation Unit 210 display unit 220 input unit 230 update unit 240 user interface

Claims (4)

Machine sensor data acquired by multiple sensors,
A state monitoring device that diagnoses a machine abnormality based on a diagnosis model that is a condition for judging a machine abnormality and displays the state of the machine,
A trigger input unit that is activated when the diagnostic model is added or updated ;
A storage unit that stores the diagnostic model of the plurality of machines,
A search unit for retrieving similar cases from the storage unit based on a diagnosis model update and addition reason selection result input from the trigger input unit, an original diagnosis model, and a newly added diagnosis model ;
A temporary storage unit for storing the diagnostic model searched by the search unit, a change table and a candidate diagnostic model ;
A display unit for displaying the diagnostic model reflecting the change table of the temporary storage unit and the candidate diagnostic model;
While browsing the display unit, the user inputs a change unit of a diagnostic model to be added or updated , and an input unit that selects a candidate diagnostic model of the temporary storage unit ;
An evaluation unit for verifying a new diagnostic model reflecting the candidate diagnostic model ;
A state monitoring apparatus comprising: an update unit that writes a new diagnostic model of the temporary storage unit to the storage unit based on a result of the evaluation unit. In claim 1,
The diagnostic model includes header information including a diagnostic model ID, sensor information, algorithm information, model information including pre-processing information and post-processing information applied before and after the algorithm, and a diagnostic model described in the model information A state monitoring apparatus comprising verification result information representing a result verified using a computer. In claim 2,
A change portion determination unit is provided that prioritizes a change unit of a diagnosis model from similar diagnosis models stored in the temporary storage unit and reflects the change in a change table stored in the temporary storage unit. Condition monitoring device. In claim 2,
A state monitoring apparatus comprising a candidate generation unit that creates a candidate for a diagnostic model by combining information of a change unit of a similar diagnostic model using information of a change table stored in the temporary storage unit .